Projects

Overview: pagmo2 is a C++ library designed for parallel optimization tasks, providing a unified interface for algorithms and problems, making it easy to deploy in massively parallel environments. Key features: - Supports both global and local optimization tasks - Utilizes the asynchronous generalized island model for efficient parallel computation - Integrates seamlessly with other optimization algorithms and problem formulations Tech stack: C++ Why it matters: pagmo2 simplifies the process of implementing and deploying parallel optimization algorithms, making it an essential tool for researchers, educators, and practitioners in the field of computational optimization.

Overview: **unfake.js** is a JavaScript library designed to clean up AI-generated pixel art and vector images, making them usable in a browser environment. Key features: - **Pixel Art Processor**: Automatically cleans up pixel art with intelligent scale detection, content-aware downscaling, grid snapping, color quantization, and artifact cleanup. - **Image Vectorizer**: Converts raster images into clean, scalable SVG files with noise reduction, smart color quantization, background handling, post-quantization smoothing, and fine-tuned tracing. Tech stack: JavaScript Why it matters: **unfake.js** simplifies the process of working with AI-generated pixel art and vector images, making them more usable and accessible in a browser environment.

Overview: This repository contains the full source code for the Claude project, providing a comprehensive view of the project's development and implementation. Key features: The repository includes the full source code, allowing developers to access and understand the project's development process. Tech stack: Python, JavaScript, and various libraries and frameworks. Why it matters: This repository is essential for developers looking to understand and contribute to the Claude project, providing a detailed look at the project's implementation and development process.

Overview: FacTool is a tool designed to detect factual errors in texts generated by large language models (LLMs), including ChatGPT. It supports four key tasks: knowledge-based QA, code generation, mathematical reasoning, and scientific literature review. Key features: - Supports knowledge-based QA, code generation, mathematical reasoning, and scientific literature review. - Includes a plugin for ChatGPT. - Available in multiple languages including English, Chinese, and Japanese. Tech stack: Python. Why it matters: FacTool is crucial for ensuring the accuracy of information generated by LLMs, enhancing the reliability of AI-generated content across various domains. Repository: GAIR-NLP/factool Primary language: Python GitHub stars: 936 GitHub description: FacTool: Factuality Detection in Generative AI GitHub URL: https://github.com/GAIR-NLP/factool

### Overview: gtrk 是为 AI Agent 设计的视频创作 CLI，让你的创意与执行无缝衔接。只需一句话，gtrk 就能将口播、访谈、稿子、画面和字幕整合成一个完整的视频工程。 ### Key features: - **口播剪辑**：将口播视频转化为可精修的工程。 - **解说制作**：从文字稿到解说视频的全流程。 - **动态图与配乐**：为视频添加动态效果和背景音乐。 - **字幕叠加**：在视频中加入字幕以增强信息传达。 ### Tech stack: TypeScript ### Why it matters: gtrk 是一个简单易用的工具，让你的创意与执行无缝衔接，只需一句话就能将口播、访谈、稿子、画面和字幕整合成一个完整的视频工程。无论你是初学者还是专业创作者，都能轻松上手。

Overview: This repository contains a collection of quizzes and assignments from Coursera courses, designed to assist learners facing difficulties in their learning process. Key features: - [Bayesian_Statistics_From_Concept_to_Data_Analysis](./Bayesian_Statistics_From_Concept_to_Data_Analysis_UC_Santa_Cruz) - [Learn_to_Program_Crafting_Quality_Code_University_of_Toronto] - [Neural_Networks_for_Machine_Learning_University_of_Toronto] - [Specialization_Applied_Data_Science_with_Python_University_of_Michigan] - [Specialization_Big_Data_UC_San_Diego] - [Specialization_Data_Mining_UIUC] - [Specialization_Data_Science_Johns_Hopkins_University] - [Specialization_Deep_Learning_deeplearning.ai] - [Specialization_Functional_Programming_in_Scala] - [Specialization_Fundamentals_of_Computing_Rice_University] - [Specialization_Machine_Learning_University_of_Washington] - [Specialization_Probabilistic_Graphical_Models_Stanford_University] - [Specialization_Program_Design_Algorithm_Peking_University] - [Specialization_Recommender_System_University_of_Minnesota] - [Specialization_Statistics_with_R_Duke_University] Tech stack: Jupyter Notebook Why it matters: This repository serves as a valuable resource for learners who need additional support in their Coursera courses, helping them better understand and apply the concepts covered

### Overview: ChatLLM.cpp is a pure C++ implementation of several models designed for real-time chatting on your computer. It supports inference of models ranging from less than 1B to more than 300B, enabling high-quality, real-time multimodal chat with RAG technology. ### Key Features: - **Model Inference**: Supports inference of models ranging from less than 1B to more than 300B. - **Multimodal Chat**: Real-time chat with support for multimodal interactions. - **RAG Integration**: Integrates with RAG technology for enhanced chat experiences. - **Pure C++**: Written entirely in C++, ensuring high performance and efficiency. - **GPU Acceleration**: Utilizes GPU acceleration for faster processing times. - **Reversed Role**: Demonstrates a reversed role feature. - **Continued Generation**: Supports continued generation for more natural conversations. - **Generation Steering**: Allows for generation steering for better chat outcomes. - **Tool Calling**: Enables tool calling for enhanced chat functionality. ### Tech Stack: - **Language**: C++ ### Why It Matters: ChatLLM.cpp offers a pure C++ implementation of models that can deliver accurate or better results than other implementations, making it a valuable tool for developers looking to enhance their chatbot capabilities. The integration with RAG technology and GPU acceleration ensures high performance and real-time chat experiences. ### Repository: - **Repository**: foldl/chatllm.cpp - **GitHub Stars**: 933 - **GitHub Description**: Pure C++ implementation of several models for real-time

Overview: This project utilizes OpenAI's CLIP neural network to search inside YouTube videos, enabling users to find relevant images by natural language queries. Key features: - Natural language search inside YouTube videos - Integrated with Huggingface Spaces and Gradio for easy access on Google Colab - Supports various queries such as "A fire truck", "Road works", "People crossing the street", "The Embarcadero", "Waiting at the red light", "Green bike lane", "A street with tram tracks", and "The Transamerica Pyramid" Tech stack: Jupyter Notebook Why it matters: This project provides a powerful tool for anyone looking to explore and analyze YouTube videos using natural language queries, making it easier to find relevant images and information.

Overview: This repository provides an introduction to Machine Learning and Artificial Intelligence, covering fundamental concepts, algorithms, and applications. Key features: - Machine Learning Fundamentals - Artificial Intelligence Concepts - Hands-on Projects Tech stack: Python, TensorFlow, PyTorch Why it matters: Understanding Machine Learning and Artificial Intelligence is crucial for anyone interested in technology and innovation.

### Overview: ViSP (Visual Servoing Platform) is an open-source software framework designed for computer vision and robotics applications. It provides a comprehensive set of tools and algorithms for visual servoing, object tracking, and image processing, making it a valuable resource for developers and researchers in the field. ### Key Features: - **Visual Servoing**: Enables precise control of robotic systems through visual feedback. - **Object Tracking**: Advanced algorithms for real-time tracking of objects in videos. - **Image Processing**: Comprehensive set of image processing tools for feature detection and analysis. ### Tech Stack: C++ ### Why It Matters: ViSP is essential for anyone working on robotics, computer vision, or image processing projects. Its open-source nature and extensive feature set make it a powerful tool for developers looking to enhance their projects with advanced visual servoing capabilities and robust image processing techniques. ### GitHub Stars: 933 ### Repository: [lagadic/visp](https://github.com/lagadic/visp) ### Primary Language: C++ ### Build Status: - Ubuntu 20.04, 22.04 (amd64): [](https://ci.appveyor.com/project/fspindle/visp/branch/master) - macOS 13 and 14: [![macos workflow](https://github.com/lagadic/visp/actions/workflows/macos.yml/badge

### Overview: DeterminFlow 是一个面向生产的 AI 工作流运行框架，旨在快速开发、验证和恢复复杂 AI 工作流，并将其稳定交付为服务。 ### Key features: - **Workflow 编排：** 可视化 Workflow Editor，支持变量、条件、并行、循环、人工审批和子流程。 - **可靠执行：** 自动重试、人工重试、跳过，以及从失败节点恢复。 - **LLM 运行边界：** 每个 Agent Node 都有独立会话和 Token 账本，工具白名单、黑名单、Workspace 和最大轮次可以按节点配置。 - **观察与交付：** 按 Workflow、Task、Node、尝试和模型调用查看状态与用量，提供 FastAPI、React 控制台。 ### Tech stack: Python ### Why it matters: DeterminFlow 已经在笔枢写作的 AI 小说生产链路中完成生产验证，显著提高了开发效率、维护便利性和流程稳定性，同时显著节省了 Token 消耗，特别是在长流程中。

### Overview: Stealing Ur Feelings is an innovative AI-powered AR experience that uses facial recognition to reveal the dangers of Big Tech's emotional surveillance programs. The project, which won the 2020 Webby Award for Best Immersive Documentary, analyzes your facial reactions to reveal your deepest secrets. ### Key Features: - **Facial Recognition**: Analyzes your facial expressions to reveal hidden emotions. - **AI-Powered**: Uses advanced AI techniques from corporate patents. - **Interactive Experience**: Immersive and engaging AR experience. ### Tech Stack: JavaScript ### Why It Matters: Stealing Ur Feelings is a groundbreaking project that sheds light on the ethical implications of Big Tech's emotional surveillance programs. By using facial recognition to reveal hidden emotions, the project challenges the notion of privacy and the potential misuse of AI in surveillance. The project's success in winning the Webby Award underscores its impact on the digital landscape and its ability to provoke thought and discussion about AI ethics. Repository: noahlevenson/stealing-ur-feelings GitHub Stars: 932 GitHub Description: Winner of Mozilla's $50,000 prize for AI

### Overview: This repository is a curated list of Sentiment Analysis methods, implementations, and resources, focusing on the sentiment analysis of written languages. It serves as a valuable resource for researchers and developers interested in the field, providing links to scholarly papers, open-source implementations, and APIs. ### Key Features: - **Contents**: A comprehensive list of Sentiment Analysis resources, including survey papers, baseline systems, open-source implementations, and APIs. - **Objective**: To provide adequate links for scholars and developers interested in sentiment analysis. - **Introduction**: Sentiment Analysis evaluates opinions, sentiments, and emotions from written texts. It includes document-level, sentence-level, and aspect-level approaches. - **Survey Papers**: Provides a survey of sentiment analysis and opinion mining. - **Baseline Systems**: Includes simple and bigrams baselines for sentiment and topic classification. - **Open Source Implementations**: Covers NodeJS, Java, Python, R, Golang, Ruby, and C# implementations. ### Tech Stack: None specified. ### Why it Matters: Sentiment analysis is crucial in various applications, including customer feedback analysis, social media monitoring, and market sentiment prediction. This repository helps developers integrate sentiment analysis into their projects efficiently. ### Repository: xiamx/awesome-sentiment-analysis - **GitHub Stars**: 933 - **GitHub Description**: A curated list of Sentiment Analysis methods, implementations, and misc. - **GitHub URL**: [Awesome Sentiment Analysis](https://github.com/xiamx/awesome-sentiment-analysis)

### Overview: This project focuses on vehicle detection, tracking, and counting using TensorFlow Object Counting API. It provides predictions for vehicle speed, color, and size, and saves cropped images of detected vehicles. ### Key Features: - **Vehicle Detection and Classification:** Uses TensorFlow Object Detection API for accurate detection and classification of vehicles (car, truck, bicycle, motorcycle, bus). - **Vehicle Color Recognition:** Utilizes OpenCV with K-Nearest Neighbors Machine Learning Classification Algorithm for color recognition. - **Vehicle Speed Prediction:** Uses OpenCV for image pixel manipulation and calculation to predict vehicle speed. - **Vehicle Size Prediction:** Predicts approximate vehicle size using OpenCV. - **Image Cropping:** Saves cropped images of detected vehicles in the "detected_vehicles" folder. - **Output CSV File:** Generates a .csv file containing detailed information about each detected vehicle, including type/size, color, movement direction, and speed. ### Tech Stack: - Python - TensorFlow Object Detection API - OpenCV ### Why It Matters: This project offers advanced capabilities for vehicle detection, tracking, and counting, providing valuable insights into traffic conditions and improving traffic management systems.

### Overview: Keystone-Bench is a clinical decision-evidence benchmark designed to evaluate chat assistants. It focuses on the primary action taken by the assistant in response to a given set of evidence, rather than the evidence itself or the overall rubric grading. ### Key features: - **Decision-Framed Conversations**: Each conversation includes a decision frame, a twin that changes exactly one thing, and a control that leaves the conversation unchanged. - **HealthBench Source Conversations**: The benchmark uses HealthBench conversations as the primary source material, ensuring consistency and accuracy. - **Twin Annotations**: The twins are annotated with evidence states, actions taken, and rubric applicability, providing a clear framework for evaluation. ### Tech stack: - Python ### Why it matters: Keystone-Bench provides a structured way to evaluate chat assistants by focusing on the primary action taken in response to evidence, rather than the evidence or the overall rubric grading. This approach ensures that the evaluation is specific and measurable, providing clear insights into how the assistant behaves in different scenarios.

### Overview: Papermorph is an AI skill that converts textbooks into interactive web experiences, turning static books into animated, narrated, and fully interactive lessons. It transforms textbooks into engaging, educational content that can be accessed on any web browser. ### Key Features: - **Textbooks:** Converts PDFs of textbooks into interactive lessons with step-by-step explanations, quizzes, and animations. - **Science:** Illustrates scientific concepts with paper puppets and animations. - **Stories:** Turns stories into paper-puppet films with subtitles and sound effects. - **Books:** Creates documentaries of public-domain images, paintings, and maps. ### Tech Stack: - JavaScript ### Why It Matters: Papermorph is a powerful tool for educators and learners alike, offering a dynamic and engaging way to explore and learn from textbooks and stories. It transforms static content into interactive, narrated experiences that enhance understanding and retention. ### Repository: - **Repository:** DozenTwelve/Papermorph - **GitHub Stars:** 873 - **Primary Language:** JavaScript ### Get Started: To use Papermorph in your project, follow these steps: 1. Install the Skill: ```bash npx skills add DozenTwelve/Papermorph --skill papermorph --agent claude-code ``` 2. Configure the Skill: ```text /papermorph ``` 3. Choose a Chapter: Papermorph suggests a chapter to start with based on your project's content. 4. Customize the Experience: Your choices are saved in `style.json`, ensuring consistency

Overview: DIR aims to learn from imbalanced data with continuous targets, addressing potential missing data for certain regions and generalizing to the entire target range. Key features: - Deep Imbalanced Regression (DIR) - Label distribution smoothing (LDS) - Feature distribution smoothing (FDS) Tech stack: Python Why it matters: DIR is a novel approach that can handle continuous and infinite target values, making it suitable for various real-world tasks such as age prediction, text similarity score prediction, and health condition score prediction. By addressing imbalanced data, it ensures better generalization and performance across different datasets and models. Repository: YyzHarry/imbalanced-regression Primary language: Python GitHub stars: 929 GitHub description: [ICML 2021, Long Talk] Delving into Deep Imbalanced Regression README excerpt: This repository contains the implementation code for the paper "Delving into Deep Imbalanced Regression". It includes large-scale datasets for tasks such as age, text similarity score, and health condition score prediction. The code is organized into subfolders for different datasets and models, and examples are provided on how to apply Label Distribution Smoothing (LDS) and Feature Distribution Smoothing (FDS) on other datasets and models.

Overview: This project presents a novel method for surface normal estimation that addresses the limitations of general-purpose dense prediction models by utilizing per-pixel ray direction and learning the relationship between neighboring surface normals. The proposed approach generates crisp, yet piecewise smooth predictions, showcasing improved generalization ability compared to state-of-the-art methods. Key features: - Utilizes per-pixel ray direction - Encodes the relationship between neighboring surface normals by learning their relative rotation - Generates crisp predictions for challenging in-the-wild images of arbitrary resolution and aspect ratio Tech stack: Jupyter Notebook Why it matters: The method addresses the need for accurate surface normal estimation in challenging real-world scenarios, providing better generalization and performance compared to existing techniques.

Overview: Citadel is an open-source operating layer designed to enhance the capabilities of Claude Code and OpenAI Codex, providing persistent project memory, intent routing, safety hooks, cost telemetry, and parallel agent fleets. This project aims to make the integration of these powerful AI tools more seamless and secure. Key features: - Persistent project memory - Intent routing - Safety hooks - Cost telemetry - Parallel agent fleets Tech stack: JavaScript Why it matters: Citadel offers a robust solution for integrating Claude Code and OpenAI Codex, ensuring that the project's state is preserved across sessions and that the AI tools can be used effectively and safely. Repository: GitHub - SethGammon/Citadel GitHub stars: 927 GitHub description: The operating layer for Claude Code + OpenAI Codex: persistent project memory, intent routing, safety hooks, cost telemetry, and parallel agent fleets.

### Overview: CLTK is a Python library designed for natural language processing (NLP) of pre-modern languages. ### Key features: - **OpenAI GenAI Backend**: Supports advanced annotation and machine learning models. - **Stanza Backend**: Provides discriminative NLP capabilities using Stanford Stanza. - **Local LLM Support**: Utilizes local language models for more efficient processing. - **Installation**: Easy to install with `pip install cltk` or additional extras for more features. ### Tech stack: Python ### Why it matters: CLTK offers a comprehensive solution for NLP tasks in pre-modern languages, addressing the unique challenges and needs of these historical texts. ### Repository: cltk/cltk ### GitHub stars: 925 ### Installation: ```bash pip install cltk ``` ### Optional Extras: - **GenAI (OpenAI-backed annotation)**: Enhances annotation capabilities. - **Stanza (discriminative NLP backends powered by Stanford Stanza)**: Improves NLP performance. - **Local LLMs via Ollama**: Supports local language models for faster processing. - **Ollama Cloud**: Access to hosted Ollama models for advanced NLP tasks. ### Choosing a Model: - **OpenAI Backend**: Uses models like "gpt-5-mini" and "gpt-5". - **Local LLMs via Ollama**: Supports models like "llama3.1:8b" and "qwen2.5:14b". ### Citation: When